ddiruscio/logseq
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[logseq-plugin-git:commit] 2025-06-05T08:36:10.944Z

Browse Source
This commit is contained in:
Davide Di Ruscio
2025-06-05 10:36:12 +02:00
parent b2116c22df
commit e3725ac9e7
6304 changed files with 708459 additions and 464 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pages/SATToSE_2023_paper_5.md
+39
View File
@@ -0,0 +1,39 @@
type:: [[REVIEWS]]
tags:: [[Image Classification]] [[AI]]
year:: 2023
venue:: [[SATTOSE]]
full-title:: How May Deep Learning Testing Inform Model Generalizability? The Case of Image Classification
date-start:: [[04-05-2023]] - 18:25
date-submitted::
external-links::
status:: [[DONE]]
file:: ![SATToSE_2023_paper_5.pdf](../assets/SATToSE_2023_paper_5_1683217643397_0.pdf)
- [[Highlights]]
- ((6453e03f-e76a-470f-8a2b-f9980b90a56f))
- I think this is more a problem related to the quality of data, isn't it?
- ((6453e278-5ca7-4a88-8045-0faefbd45f0d))
- ((6453e326-7ee7-44c9-a95d-7223e809e6d3))
- Why have you developed yet another neural network for image classification? To assess the generalibility of your results, you should consider an existing approach.
- Why not considering one of the existing baselines?
- ((6453e3cd-e8c4-40dd-a28c-26578f1e411d))
- This is a potential [[source]] of [[bias]] because noises that make sense to introduce can depend on the application domain / goal of the model under analysis
- ((6453e437-a1c2-478a-883b-d245eaed4d69))
- Why do you need to answer such a research question?
-
- [[Comments]]
- The [[paper]] presents a preliminary analysis of how deep learning solutions can decrease their accuracy when put at [[WORK]] in real conditions. The analysis has been done [[by]] considering the image classification case. A CNN model has been developed and its performance resulted to be over 90%. When applied with different data, the overall performance decreased substantially.
- The [[paper]] is about an important problem. Overall, the [[paper]] is well-written and well-structured. However, I have the following concerns about RQ1 of the [[paper]]:
- It is not clear why the [[authors]] decided to implement yet another CNN model for the image classification problem. This is a potential [[bias]] and a potential threat to the validity of the performed analysis
- The need of RQ1 is not clear to me. In other words, why the [[authors]] considered the need of corroborating previous findings in the field of image recognition?
- RQ2 is very important and needs further investigation. It is not clear if the [[source]] of the problems is the quality of the training/testing data, on the validation procedures that are adopted, both ... Thus, in this respect, I think the [[paper]] is very interesting for the event.
- RIVISTA
- The [[paper]] presents a preliminary analysis of how deep learning solutions can decrease their accuracy when put at [[WORK]] in real conditions. The analysis has been done [[by]] considering the image classification case. A CNN model has been developed, and its performance was over 90%. When applied with different data, the overall performance decreased substantially.
- The paper is about an important problem. Overall, the [[paper]] is well-written and well-structured. However, I have the following concerns about RQ1 of the paper:
- The authors should explain their decision to implement a new CNN model (instead of using existing ones) for image classification to avoid potential bias and threats to the validity of the analysis.
- The need for RQ1 needs to be clarified for me. In other words, why did the [[authors]] consider the need to corroborate previous findings in the field of image recognition?
- Further investigation is required for RQ2 as it is not clear whether the problems stem from the quality of training/testing data, validation procedures, or both. Therefore, I find the paper to be quite intriguing in this regard and worth discussing at the event.
-
-
-
-